Remember When We Predicted AI Agents?

Welcome back apprentices! 👋

Remember when we predicted that 2025 would be the "Year of AI Agents"? 

Well, turns out, reality is catching up fast. OpenAI just handed businesses a toolkit for building AI employees, and these bots don’t just chat — they actually work. 

AI is no longer just that overeager assistant who suggests dinner recipes when you’re clearly searching for tax deductions — it’s stepping into the workplace, tackling real tasks, and, unlike Steve from accounting, never "forgets" to reply to emails.

So the real question is: Are we about to witness the best productivity boost of our lifetime… or the moment Clippy finally gets revenge?

In today's email

• McDonald's Is Going Full AI

• Meta’s DIY AI Chips: Bye Bye, Nvidia?

• AI Just Passed Peer Review

• AI Stock Watch (New)

• Is Your Next Employee an AI?

• Say Hello to AI 

• The New Gemini Robotics

_{Read Time: 4 minutes}

Quick News 

🍟 McDonald’s is deploying AI across 43,000 locations to boost efficiency, improve order accuracy, and (hopefully) stop ice cream machine breakdowns. Partnering with Google Cloud, it's introducing real-time AI for predictive maintenance, computer vision for order accuracy, and a generative AI "virtual manager" to help employees juggle the chaos of drive-thru, delivery, and dine-in orders.

💾 Meta is officially testing its first in-house AI training chip, hoping to stop handing Nvidia its entire paycheck and shave billions off its $65B AI infrastructure bill. Built by TSMC and part of the MTIA series, it’s designed for AI training and inference workloads — aka, making AI better at recommending cat videos. If all goes well, Meta plans to ditch more third-party chips and deploy these at scale by 2026. 

📜 Japanese AI startup Sakana built an AI that wrote, analyzed, and visualized an entire research paper, and — somehow — it got accepted at ICLR 2025, scoring higher than some human submissions. Sure, it messed up citations (relatable), but it’s the first fully AI-generated paper to pass peer review — which means we might soon be grading AI’s homework instead of the other way around.

Together with Superhuman AI

Start learning AI in 2025

Keeping up with AI is hard – we get it!

That’s why over 1M professionals read Superhuman AI to stay ahead.

• Get daily AI news, tools, and tutorials

• Learn new AI skills you can use at work in 3 mins a day

• Become 10X more productive

Sign up and start mastering AI

OpenAI
Is Your Next Employee an AI? OpenAI Thinks So

OpenAI just dropped a toolkit for building AI agents, meaning businesses can now create custom bots that actually get work done — not just reply with “I’m sorry, I can’t do that.” 

The new Responses API gives AI agents web search, file management, and computer-use superpowers, while an open-source Agents SDK lets developers build single or multi-agent systems (yes, plural — because one AI isn’t enough anymore). The old Assistants API is being retired in 2026, so consider this the next evolution of AI workers.

What’s the Deal? 2025 is officially the “Year of AI Agents” — and this time, they’re not just hype. Stripe’s AI is already handling invoicing, Box’s AI is searching through enterprise docs, and somewhere, a human intern just sighed in relief. OpenAI is closing the gap between flashy demos and actual workplace automation, making AI assistants that don’t just chat, but actually work.

Your next co-worker might be an AI that never complains, never takes PTO, and never forgets to attach the file. Welcome to the future — just don’t expect it to chip in for coffee.

Sesame
Say Hello to AI That Actually Sounds Human

Sesame, the AI startup that made waves with its eerily natural-sounding assistant Maya, just cracked open the code vault and released its first open-source base model, CSM-1B. The model, designed for real-time, context-aware voice interactions, combines a Llama-based foundation with a speech generation system that runs on Mimi audio codes. 

Translation? It can generate shockingly natural-sounding speech in real-time without the usual “robot trying to be human” awkwardness.

Now freely available on Hugging Face and GitHub, CSM-1B is a gift to developers eager to integrate high-quality AI voice synthesis into their apps, wearables, and virtual assistants. 

This move also hints at Sesame’s bigger ambitions — particularly in AI-powered smart glasses, where a truly intelligent assistant could finally make augmented reality feel useful, instead of just being a $3,500 way to check your emails.

What’s the Deal? AI voices are getting freakishly good, and now developers everywhere can fine-tune their own. From customer support bots to AI wearables, the line between talking to a human and a machine is getting blurrier by the day.

Google DeepMind
Smarter, Faster, and (Hopefully) Less Clumsy Robots

Ever wondered when AI would stop just answering questions and start actually doing stuff? Google DeepMind just unveiled Gemini Robotics, a next-gen AI system designed to bring robots to life — literally. 

Built on Gemini 2.0, these bots can see, hear, and interact with the physical world, handling everything from warehouse logistics to delicate object manipulation (think folding laundry without turning it into origami).

The Big Moves:

• 2-3x improvement in robotic task success rates 

• Powered by multimodal AI, meaning it can process text, images, video, and movement — no more one-trick pony bots

• Partnering with Apptronik, Boston Dynamics, and others to bring AI-powered machines into manufacturing, logistics, and beyond

What’s the Deal? With 70% of warehouses still running on manual labor, Gemini-powered robots could be the game-changer businesses have been waiting for. Whether it’s smarter automation, AI-driven assistants, or a future where your barista is a robot — one thing’s clear: AI isn’t just talking anymore, it’s moving.

Even Quicker News 

🤝 Manus just partnered with Alibaba’s Qwen to bring AI agents to China, proving that viral AI hype can actually lead to real deals. With Qwen’s open-source models and Manus’ agent skills, this could be the ultimate AI tag team. 

📚France just hit Meta with a lawsuit, accusing it of treating copyrighted books like a free buffet for AI training. Looks like “Zuckerberg’s Library” might have some overdue fines.

💼 Salesforce is dropping $1B in Singapore to supercharge AI adoption — because apparently, digital employees are the future of the workforce.  This follows its recent $500M bets in Saudi Arabia and Argentina.

Today’s Toolbox

⚡ Google just dropped Gemma 3, an AI model so efficient it runs on a single GPU while outperforming bigger rivals like Llama-405B and DeepSeek-V3. With multimodal smarts, a 128K token memory, and support for 140 languages, it’s high-powered AI without the server farm.

🎮 Xbox’s new Copilot for Gaming, powered by Microsoft, is like an AI sidekick that helps you find games, master tricky levels, and avoid embarrassing Google searches like “how to beat the first boss.” Now available in preview for Xbox Insiders. 

🚀 Foxconn just launched FoxBrain, an AI model built in a month using 120 Nvidia H100 GPUs — because making iPhones wasn’t enough. If suppliers are building AI this fast, what’s Apple’s excuse?

Let's Chat About AI

This is all for today!

Replay to this email with your thoughts.

AI is more than just a buzzword. It’s a shift in how we live and work. And understanding it a bit better means you can make smarter choices about the tech you use every day.

_FEEDBACK